Early detection of arthritis, particularly rheumatoid arthritis, is extremely important to prevent joint damage but currently existing early-stage detection methods are insensitive at detecting the condition. This paper presents a Dynamic Multimodal Attention-Integrated Hybrid Ensemble (DMAIHE) framework that combines clinical, biomarker, imaging and wearable sensor modalities to effectively achieve accurate and interpretable predictions of arthritis. Each modality is processed independently through its own windowed subnetwork: LightGBM for clinical data and biomarkers, ResNet-50 for imaging data, and LSTM for wearable data, and then is fused together using dynamic attention that weighs each modality independently for each patient's context. Experiments utilizing a synthetically created multi-modal arthritis dataset of 2000 patients demonstrated and the model output performed the existing models outperformed resulting performance with ROC-AUC, Accuracy, Recall 0.921, 0.96, 0.90 respectively.